Estimating Calorie Expenditure from Output Voltage ofPiezoelectric Energy Harvester - an Experimental

Feasibility Study

Guohao Lan1,2, Sara Khalifa1,2, Mahbub Hassan1,2, and Wen Hu1,2

1School of Computer Science and Engineering, University of New South Wales, Sydney, NSW, Australia

2National ICT Australia, Alexandria, NSW, Australia

{glan, sarak, mahbub, wenh}@cse.unsw.edu.au

ABSTRACTThere is a growing interest in developing energy harvest-ing solutions for wearable devices so they can self-powerthemselves without relying on batteries. Piezoelectric en-ergy harvesters (PEHs) can convert kinetic energy releasedfrom human activities into usable electrical energy for pow-ering various electronic circuits inside the wearable device.Intuitively, the kinetic energy is produced because the userexpends some calories during the physical activities. Wetherefore postulate that the voltage output of a PEH in awearable device should contain information that can be usedto estimate the amount of calorie expended. If this is true,then the PEH can be used as a new source for calorie estima-tion. Unlike conventional sensors, such as accelerometers, aPEH does not consume any power, which would make thisnew source very attractive. In this paper, using real PEHhardware and the data collected from ten real subjects, weconduct an experimental study to assess the suitability ofPEH voltage in estimating calorie expenditure for two differ-ent activities, walking and running. We find that, for mostsubjects, the calorie estimations obtained from the outputvoltage of PEH is very close to those obtained from a 3-axialaccelerometer.

Categories and Subject DescriptorsJ.3 [Computer Applications]: Life and medical sciences

General TermsExperimentation

KeywordsPiezoelectric Energy Harvester, Accelerometer, Calorie Ex-penditure Estimation

1. INTRODUCTIONCalorie expenditure estimation (CEE) is valuable in moni-toring many health problems, such as obesity, an epidemicwhich is predicted to be the most preventive health problemin the future [25]. A large number of works in the literaturehave been devoted to estimate the calorie expenditure basedon the output signals of accelerometers [3, 4, 5, 16] availablein wearable devices. By combining acceleration data withanthropometric features, such as weight, height, age, etc.,these algorithms can accurately estimate calorie expendi-ture of a user. These advancements have resulted in manypopular wearable products in the market, such as Fitbit,Apple Watch, and Jawbone, that can collect and analysehuman acceleration to estimate, display, or upload calorieexpenditure on a continuous basis.

Accelerometers consume power. At the moment, most wear-able devices are powered by batteries. There is, however, agrowing interest in developing energy harvesting solutionsfor wearable devices so they can self-power themselves with-out relying on batteries [8, 17, 23, 24, 26]. Although thereare several choices for energy harvesting, such as solar [10],wind [19], thermal [21], kinetic, etc., piezoelectric energyharvester (PEH) is known to generate the most significantpower by converting human motion to electricity [17]. IfPEH is used, then it can be used to power the accelerome-ters and other circuits in a wearable device.

In this paper we seek to find a new use of PEH for CEE.Intuitively, the kinetic energy is produced because the userexpends some calories during the physical activities. Wetherefore postulate that the voltage output of a PEH in awearable device should contain information that can be usedto estimate the amount of calorie expended. If this is true,then the PEH can be used as a new source for calorie esti-mation. Unlike conventional sensors used in CEE, such asaccelerometers, a PEH does not consume any power, whichwould make this new source very attractive.

Using real PEH hardware and the data collected from tenreal subjects, we conduct an experimental study to assessthe suitability of PEH voltage in estimating calorie expen-diture for two different activities, walking and running. Wefind that, for most subjects, the calorie estimations obtained

Human Activity Acceleromter

Anthropometric

Features

Regression

Model

Estimated

Calorie

Expenditure

accelerometer signal

weight, height, age, etc.

Figure 1: Architecture of the accelerometer basedcalorie expenditure estimation.

from the output voltage of PEH is very close to those ob-tained from a 3-axial accelerometer. The main contributionsof this paper can be summarized as follows:

• We propose the use of output voltage of PEH as a newsource of realising CEE for kinetic-powered devices,which is the first of its kind in the CEE literature.

• We propose a linear regression model based on PEHvoltage data collected from ten subjects and show that,for most subjects, the calorie estimations obtained fromthe output voltage of PEH is very close to those ob-tained from a 3-axial accelerometer.

The rest of the paper is organised as follows. In Section 2,we explain the accelerometer-based CEE method we usedin this study for comparison. We present our experimentalhardware, our data collection and the proposed PEH-basedCEE method in Section 3, followed by the performance eval-uation in Section 4. Related work is reviewed in Section 5.We conclude our work in Section 6.

2. ACCELEROMETER-BASED CEEAccelerometer based calorie expenditure estimation highlyrelies on the accelerometers to frequently sample the humanactivities. The architecture of the conventional accelerom-eter based CEE is shown in Figure 1. In order to estimatethe calorie expenditure, accelerometer outputs are combinedwith the anthropometric features of the subjects to performregression analysis [4, 5].

In this paper, we apply the popular method proposed in [4]to estimate the calorie expenditure base on the signal froma 3-axis accelerometer and the anthropometric features ofthe subjects. The regression model is given as follows:

CEEacc = a×H(k) + b× V (k) (1)

where, CEEacc represents the estimated calorie expenditureat the kth minute. H(k) was defined as the square root ofthe sum of squared of the accelerometer signals on the x andy-axes (H(k) =

√Acc2x +Acc2y). V (k) was defined as the

accelerometer signal on the z -axe (V (k) =√Acc2z). Parame-

ters a and b represent the coefficients of the regression model(1), and are generalized by stepwise multiple-linear regres-sion based on the anthropometric features of 125 subjects in

Piezoelectric

Energy Harvester

Accelerometer

Microcontroller SD Memory Card

12.5 cm

6 cm

4 cm

Figure 2: The hardware platform.

study [4]:

a =5.78×Mass+ 11.95×Height+ 6.89×Age− 2001

1000(2)

b =5.96×Mass+ 349.5

1000(3)

It has been reported that this regression model can achieve60-95% correlation [4, 18] in estimating the calorie expendi-ture for walking and running.

3. PEH-BASED CEEPizoelectric energy harvester, which harvests energy fromthe vibration generated by the human motion, has been re-garded as a promising energy source to power future wear-able sensors. Unfortunately, a recent study [14] has shownthat the amount of power that can be harvested from theenergy harvester through the body motion can hardly be suf-ficient to power the accelerometers at a high sampling rate,thus, it results in poor accuracy for the upper layer appli-cations, such as activity recognition and calorie expenditureestimation.

Instead of using the signals from the accelerometers to esti-mate the calorie expenditure, or using the energy harvestedby the energy harvester to power the accelerometer, the pri-mary idea of our method is to develop an regression modelbased on the output voltage signals from the pizoelectricenergy harvester directly. Prior to presenting the proposedmethod, we introduce the hardware we designed for thisstudy and our data collection method.

3.1 Hardware setupWe designed and built a data logger for this study. The datalogger includes a vibration energy harvesting product fromthe MIDE Technology called Volture, which implements thetransducer to provide AC voltage as its output. Our hard-ware also includes a 3-axis accelerometer (MMA7361LC)to record the acceleration signals, simultaneously with thevoltage signal. An Arduino Uno has been used as a micro-controller device for sampling the data from both the Voltureand the accelerometer. A sampling rate of 1 KHz has beenused for data collection. The sampled data has been savedon an 8GB microSD card which has been equipped to theArduino using microSD shield. A 9 volts battery has been

Piezoelectric Energy

Harvester (Volture V25W)

3-axis Accelerometer

(MMA7361LC)

MicroSD Shield Arduino Uno

Microcontroller

8 GB MicroSD Card

9 Volt Battery to

power Arduino

2 switches to switch on/off the device and

control the start/stop data logging

Figure 3: The internal appearance of the data log-ger.

used to power the Arduino. To control the data collection,our data logger also includes two switches, one to switchon/off the device and the other to control the start and stopof data logging. The hardware platform and the internalappearance of the data logger are shown in Figure 2 andFigure 3, respectively.

3.2 Data collectionTen healthy subjects (4 male, and 6 female) of different eth-nic backgrounds from our lab volunteered to participate inthis research study. The participants were asked to hold thedata logger in either their left or right hand and perform twodifferent activities: walking and running. The details of theanthropometric features of the subjects are as follows: Age(26-35 years, µ = 29, and σ = 3.06), Weight (58-91 Kg, µ =69.3, and σ = 10.21), Height (154-185 cm, µ = 168.5, and σ= 9.98).

All subjects performed walking and running with their nat-ural speed, i.e., there was no special speed control, thus, thestep frequencies of walking and running vary among differ-ent subjects. To avoid mislabeling, the switch on the datalogger has been used to control the start and stop of datacollection at the beginning and end of each activity. Sub-jects were asked to stop and wait a few seconds after anactivity and before starting the next activity. The data col-lected between the start and stop times of an activity werelabeled with the name of that activity. Each subject pro-vided between 25 and 35 seconds of data for both walkingand running.

3.3 Proposed Calorie Expenditure Estimationusing Harvested Voltage Signal

The architecture of the proposed pizoelectric energy har-vester based method is shown in Figure 4. The system con-tains a kinetic energy harvester which harvests energy di-rectly from human activity. Instead of using the harvestedenergy to power the accelerometer, the output voltage isused as the input signal, together with the subject’s anthro-pometric features, to generate a regression model for calorieexpenditure. No acceleration data is used in the proposedarchitecture comparing with the accelerometer based archi-tecture given in Figure 1.

Human ActivityEnergy

Harvester

Anthropometric

Features

Regression

Model

Estimated

Calorie

Expenditure

voltage signal

weight, height, age, etc.

Figure 4: Architecture of the PEH based calorie ex-penditure estimation.

3.4 Regression ModelWe performed a regression analysis to generate the general-ized regression model. The linear regression model can berepresented as follows:

CEEvolt = Xβ + ε (4)

where, CEEvolt indicates the estimated calorie expenditureat the kth minute. X denotes the vector of input signals, in-cluding the anthropometric features of the subjects (weight,height, and age), and the output voltage signals from theenergy harvester. The β and ε are the vector of coefficientsand residual error, respectively.

Without the ground truth from the indirect calorimeter, weapplied the accelerometer based method proposed given inEquation (1) to estimate the calorie expenditure of the col-lected activity traces. The estimated results are used as theground truth to train and calibrate our regression model.Note that, because Equation (1) is a generalized model, ourregression model will not specify the activity type.

For each subject, Si, in our subjects set, we have collectedtwo acceleration signal traces, one for walking, and one forrunning. Using those two signal traces as inputs, we applyEquation (1) to get the acceleration based calorie expendi-ture estimations, CEEacc walk, and CEEacc run, for walkingand running activity of subject Si, respectively. These twoacceleration based CEE results are synchronized with therecorded output voltage signal from the PEH during Si’swalking and running.

As an example, Figure 5 exhibits the output signals of the3-axis accelerometer for both walking and running of sub-ject 2. Together with the anthropometric features of subject2, the acceleration signals are used to estimate the calorieexpenditure (apply Equation (1), (2), and(3)). The resultsCEEacc walk and CEEacc run of subject 2 are given in Fig-ure 6(a), which clearly indicates that running can expendmore calories than walking. Correspondingly, Figure 6(b)shows the synchronized output voltage signals of the PEHrecorded by our data logger for the same subject. Expect-edly, PEH can harvest more energy from running than walk-ing.

In order to generate the regression Equation (4), for eachsubject, Si, we used the rest 9 subjects as the training setto train the linear regression model, and applied the Leave-one-out cross-validation to evaluate the performance of the

Time (Seconds)

0 1 2 3 4 5 6 7 8 9 10

Accele

ration (

m/s

2)

-8

-6

-4

-2

0

2

4

6

X-axis Y-axis Z-axis

(a) Accelerometer Output for walking

Time (Seconds)

0 1 2 3 4 5 6 7 8 9 10

Accele

ration (

m/s

2)

-20

-15

-10

-5

0

5

10

15

X-axis Y-axis Z-axis

(b) Accelerometer Output for running

Figure 5: The output signal of the 3-axis accelerom-eter for both walking and running of subject 2.

generated regression model on Si. The Pearson’s correlationcoefficients of the linear correlation between the estimatedcalorie expenditure, CEEvolt, and the input signal vector,X, for all the ten subjects are given in Figure 7. In caseof walking, the Pearson’s r is 0.78 on average, and in caseof running, we achieved a value of 0.71 on average, whichindicates a strong positive correlation between the voltagesignals and calorie expenditure for both walking and run-ning.

4. EXPERIMENTS RESULTSThe instantaneous estimations of the PEH-based CEE andthe ACC-based CEE, of walking and running, for all the tensubjects are given in Figure 8. Every second includes 1000estimations. The results show that although the instanta-neous estimations of our proposed PEH-based method aredifferent from that of the ACC-based method, the averagesof the PEH-based CEE over a period of time (one secondor longer), are very close to that of the ACC-based meth-ods. The average value over longer terms is more usefulin practical applications, comparing with the instantaneousmeasurements for every second. In addition, Figure 9 plotsthe mean of the estimated calorie expenditure over one sec-ond of the PEH-based and ACC-based methods. Figure9(a) compares the CEEvolt of the PEH-based method withthe CEEacc of the ACC-based methods for walking. The

Time (seconds)

0 1 2 3 4 5 6 7 8 9 10

Calo

rie E

xpenditure

(K

J/m

in)

20

40

60

80

100

120 Walking Running

(a) Plot of the CEEacc walk and CEEacc run of subject 2, usingthe regression model given in Equation (1)

Time (seconds)

0 1 2 3 4 5

AC

Voltage (

Volts)

-5

0

5

10

15

20

25

Walking Running

(b) Plot of the output voltage signals of pizoelectric energy har-vester

Figure 6: Plot of the output voltage and calorie ex-penditure for subject 2.

0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

1 2 3 4 5 6 7 8 9 10

Pears

on

's r

Subject

Walking Running

Figure 7: The Pearson’s correlation coefficients ofthe linear correlation between the estimated resultsCEEvolt and the input signal vector X for all tensubjects.

Mean Absolute Percentage Error (MAPE) is 0.12 in case ofwalking. Similarly, Figure 9(b) shows the results for run-ning and the MAPE is 0.16. However, which is indicatedin Figure 8(o) and Figure 9(a), the PEH-based method willoverestimate the calorie expenditure for walking. In addi-tion, Figure 8(f)(l)(t) expose that the PEH-based method

Time (seconds)

0 5 10 15 20 25

KJ/m

in

50

60

70

80

PEH-based CEE

ACC-based CEE

(a) subject1:walking

Time (seconds)

0 1 2 3 4 5 6 7

KJ/m

in

0

50

100

150 PEH-based CEE

ACC-based CEE

(b) subject1:running

Time (seconds)

0 2 4 6 8

KJ/m

in

30

40

50

60

70

PEH-based CEE

ACC-based CEE

(c) subject2:walking

Time (seconds)

0 2 4 5 8 10

KJ/m

in

0

50

100

150

PEH-based CEE

ACC-based CEE

(d) subject2:running

Time (seconds)

0 2 4 6 8 10 12 14 16 18

KJ/m

in

40

60

80

100

120PEH-based CEE

ACC-based CEE

(e) subject3:walking

Time (seconds)

0 1 2 3 4 5 6

KJ/m

in

0

50

100

150

200

250

PEH-based CEE

ACC-based CEE

(f) subject3:running

Time (seconds)

0 2 4 6 8

KJ/m

in

30

40

50

60

70

80

PEH-based CEE

ACC-based CEE

(g) subject4:walking

Time (seconds)

0 1 2 3 4 5 6 7 8 9

KJ/m

in

20

40

60

80

100

120

PEH-based CEE

ACC-based CEE

(h) subject4:running

Time (seconds)

0 2 4 6 8

KJ/m

in

0

50

100

PEH-based CEE

ACC-based CEE

(i) subject5:walking

Time (seconds)

0 2 4 6 8 10

KJ/m

in

0

50

100

150

PEH-based CEE

ACC-based CEE

(j) subject5:running

Time (seconds)

0 2 4 6 8

KJ/m

in

60

80

100

120

140

160

PEH-based CEE

ACC-based CEE

(k) subject6:walking

Time (seconds)

0 1 2 3 4 5K

J/m

in

0

100

200

300

PEH-based CEE

ACC-based CEE

(l) subject6:running

milisecond

0 2 4 6 8

KJ/m

in

0

20

40

60

80

PEH-based CEE

ACC-based CEE

(m) subject7:walking

Time (seconds)

0 2 4 6 8 10

KJ/m

in

0

50

100

150

PEH-based CEE

ACC-based CEE

(n) subject7:running

Time (seconds)

0 2 4 6 8

KJ/m

in

20

40

60

80

100

PEH-based CEE

ACC-based CEE

(o) subject8:walking

Time (seconds)

0 1 2 3 4 5 6 7 8

KJ/m

in

0

50

100

150

200

PEH-based CEE

ACC-based CEE

(p) subject8:running

Time (seconds)

0 5 10 15 20 25

KJ/m

in

30

40

50

60

70

PEH-based CEE

ACC-based CEE

(q) subject9:walking

Time (seconds)

0 2 4 6 8 10

KJ/m

in

20

40

60

80

100

120

PEH-based CEE

ACC-based CEE

(r) subject9:running

Time (seconds)

0 2 4 6 8

KJ/m

in

40

60

80

100

120

PEH-based CEE

ACC-based CEE

(s) subject10:walking

Time (seconds)

0 1 2 3 4 5 6 7 8

KJ/m

in

50

100

150

200

PEH-based CEE

ACC-based CEE

(t) subject10:running

Figure 8: The instantaneous estimation results of the PEH-based and ACC-based CEE for all the ten subjects,for both walking and running.

0

20

40

60

80

100

120

1 2 3 4 5 6 7 8 9 10

KJ/

min

Subject

Accelerometer based

Energy Harvester based

(a) Plot of the CEEvolt and CEEacc for walking

0

50

100

150

200

1 2 3 4 5 6 7 8 9 10

KJ/

min

Subject

Accelerometer based

Energy Harvester based

(b) Plot of the CEEvolt and CEEacc for running

Figure 9: The average CEE per second of the ACC-based and proposed PEH-based method.

will underestimate the calorie expenditure for running. Weinvestigated that this results from the uncontrolled intensityand step frequency of the subjects’ activity during our datacollection. We will improve this by collecting more reliabledata in our future work.

5. RELATED WORKIn the laboratory environment, the most accurate way ofcalorie expenditure estimation is to use the indirect calorime-ters. Indirect calorimeter, such as the COSMED k4b2, esti-mates the amount of energy expenditure by measuring theconsumption of oxygen and the production of carbon diox-ide. Although it has been reported that using the indirectcalorimeter can achieve a very high correlation with the ac-tual calorie expenditure [6, 20], it is impractical for everydaycalorie estimation, due to its inconvenience.

Outside the laboratory, wearable sensors are widely used asthe alternatives to capture the information of daily activ-ities and estimate the energy expenditure [1, 3, 5, 16, 18].Early studies [7, 22] used accelerometer signals to generate ageneralized regression model for the CEE of different activi-ties. However, a study [1] has shown that, due to the varietyof the activity intensities, methods using a single regressionmodel will underestimate the expenditure of the high inten-sity activities, but overestimate that of the low intensity ac-tivities. To resolve this shortcoming, algorithms for human

activity recognition (HAR), which have been widely used inthe indoor positioning and smart living [9, 11, 12, 13], havealso been applied in CEE. Recent studies have proposed theactivity-specific CEE, which first recognizes the activity, andthen generates an activity specific regression model [5], orassigns the Metabolic Equivalent of Task (MET, is a mea-surement of the calorie expenditure of a physical activity) tothat activity [1, 2]. Their results [1, 3] shown that the activ-ity specific CEE methods can greatly improve the accuracyof CEE.

Unfortunately, the shortcoming of those methods is the powerconsumption of the wearable sensors, especially the accelerom-eters, used in the activity recognition is relatively high [14,15]. To achieve continuous calorie estimation, we either needto instrument the wearable devices with large batteries, orfrequently replace the batteries. Obviously, neither of thesechoices is desirable for future body area wearable devices.Compared to the literature, our work in this paper is uniquein the sense that we do not use accelerometer at all to es-timate calorie expenditure. PEH-based CEE is a totallynew concept to the best of our knowledge. It opens thedoor to a completely new direction for calorie estimation.The fact that it can estimate calorie simply from the PEHdata makes this method more competitive than the conven-tional accelerometer-based techniques for the realisation ofbattery-less wearables.

6. CONCLUSION AND FUTURE WORKWe have conducted the first experimental study to estimatecalorie expenditure from the voltage output of PEH. Wehave used real PEH hardware and collected voltage datafrom 10 different subjects performing two different activi-ties, walking and running. We have found that, for mostsubjects, the calorie estimations obtained from the outputvoltage of PEH is very close to those obtained from a 3-axialaccelerometer. The result of this study shows the feasibilityof estimation the calorie expenditure from the output signalof PEH in kinetic-powered wearable devices, which providesopportunities to develop new wearable devices for calorieestimation.

Motivated by the results of this paper, we plan to extend thecurrent experimental study by collecting more data from alarger subject set, and studying a variety of activities withdifferent intensities, including standing, walking, running,and biking. Moreover, instead of applying the estimated re-sults from the accelerometer based method as the groundtruth, we will use the indirect calorimeters, COSMED k4b2,to measure the actual calorie expenditure and train our re-gression model. Another direction of our future work wouldbe generating activity-specific regression model for PEH basedCEE. For activity recognition, we would like to apply themethod proposed in [15] to achieve accelerometer-free activ-ity recognition.

7. ACKNOWLEDGEMENTWe would like to thank all the participants for their help inthe data collection. We also appreciate the anonymous re-viewers for their detailed comments, which helped improvingthe final version of this paper.

8. REFERENCES

[1] F. Albinali, S. Intille, W. Haskell, and M. Rosenberger.Using wearable activity type detection to improvephysical activity energy expenditure estimation. InProceedings of the 12th ACM international conferenceon Ubiquitous computing, pages 311–320. ACM, 2010.

[2] M. Altini, J. Penders, and O. Amft. Energyexpenditure estimation using wearable sensors: a newmethodology for activity-specific models. InProceedings of the conference on Wireless Health,page 1. ACM, 2012.

[3] M. Altini, J. Penders, R. Vullers, and O. Amft.Estimating energy expenditure using body-wornaccelerometers: a comparison of methods, sensorsnumber and positioning. Biomedical and HealthInformatics, IEEE Journal of, 19(1):219–226, 2015.

[4] K. Y. Chen and M. Sun. Improving energyexpenditure estimation by using a triaxialaccelerometer. Journal of applied Physiology,83(6):2112–2122, 1997.

[5] S. E. Crouter, K. G. Clowers, and D. R. Bassett. Anovel method for using accelerometer data to predictenergy expenditure. Journal of applied physiology,100(4):1324–1331, 2006.

[6] R. Duffield, B. Dawson, H. Pinnington, and P. Wong.Accuracy and reliability of a cosmed k4b 2 portablegas analysis system. Journal of Science and Medicinein Sport, 7(1):11–22, 2004.

[7] P. S. Freedson, E. Melanson, and J. Sirard.Calibration of the computer science and applications,inc. accelerometer. Medicine and science in sports andexercise, 30(5):777–781, 1998.

[8] M. Gorlatova, J. Sarik, G. Grebla, M. Cong,I. Kymissis, and G. Zussman. Movers and shakers:Kinetic energy harvesting for the internet of things. InThe 2014 ACM international conference onMeasurement and modeling of computer systems,pages 407–419. ACM, 2014.

[9] M. Hassan. A performance model of pedestrian deadreckoning with activity-based location updates. InNetworks (ICON), 2012 18th IEEE InternationalConference on, pages 64–69. IEEE, 2012.

[10] X. Jiang, J. Polastre, and D. Culler. Perpetualenvironmentally powered sensor networks. InInformation Processing in Sensor Networks, 2005.IPSN 2005. Fourth International Symposium on,pages 463–468. IEEE, 2005.

[11] S. Khalifa and M. Hassan. Evaluating mismatchprobability of activity-based map matching in indoorpositioning. In Indoor Positioning and IndoorNavigation (IPIN), 2012 International Conference on,pages 1–9. IEEE, 2012.

[12] S. Khalifa, M. Hassan, and A. Seneviratne. Adaptivepedestrian activity classification for indoor deadreckoning systems. In Indoor Positioning and IndoorNavigation (IPIN), 2013 International Conference on,pages 1–7. IEEE, 2013.

[13] S. Khalifa, M. Hassan, and A. Seneviratne. Featureselection for floor-changing activity recognition inmulti-floor pedestrian navigation. In MobileComputing and Ubiquitous Networking (ICMU), 2014Seventh International Conference on, pages 1–6.IEEE, 2014.

[14] S. Khalifa, M. Hassan, and A. Seneviratne. Pervasiveself-powered human activity recognition without theaccelerometer. In IEEE International Conference onPervasive Computing and Communications (PerCom).IEEE, 2015.

[15] S. Khalifa, M. Hassan, A. Seneviratne, and S. K. Das.Energy harvesting wearables for activity-awareservices. Internet Computing, IEEE, 20(5):2–10, 2015.

[16] J. Lester, C. Hartung, L. Pina, R. Libby, G. Borriello,and G. Duncan. Validated caloric expenditureestimation using a single body-worn sensor. InProceedings of the 11th international conference onUbiquitous computing, pages 225–234. ACM, 2009.

[17] P. D. Mitcheson, E. M. Yeatman, G. K. Rao, A. S.Holmes, and T. C. Green. Energy harvesting fromhuman and machine motion for wireless electronicdevices. Proceedings of the IEEE, 96(9):1457–1486,2008.

[18] A. Pande, Y. Zeng, A. K. Das, P. Mohapatra,S. Miyamoto, E. Seto, E. K. Henricson, and J. J. Han.Energy expenditure estimation with smartphone bodysensors. In Proceedings of the 8th InternationalConference on Body Area Networks, pages 8–14.ICST, 2013.

[19] C. Park and P. H. Chou. Ambimax: Autonomousenergy harvesting platform for multi-supply wirelesssensor nodes. In Sensor and Ad Hoc Communicationsand Networks, 2006. SECON’06. 2006 3rd AnnualIEEE Communications Society on, volume 1, pages168–177. IEEE, 2006.

[20] H. C. Pinnington, P. Wong, J. Tay, D. Green, andB. Dawson. The level of accuracy and agreement inmeasures of f e o 2, f e co 2 and vIG e between thecosmed k4b 2 portable, respiratory gas analysissystem and a metabolic cart. Journal of Science andMedicine in Sport, 4(3):324–335, 2001.

[21] M. Strasser, R. Aigner, M. Franosch, andG. Wachutka. Miniaturized thermoelectric generatorsbased on poly-si and poly-sige surface micromachining.Sensors and Actuators A: Physical, 97:535–542, 2002.

[22] A. M. Swartz, S. J. Strath, D. R. Bassett, W. L.O Brien, G. A. King, and B. E. Ainsworth. Estimationof energy expenditure using csa accelerometers at hipand wrist sites. Medicine and Science in Sports andExercise, 32(9; SUPP/1):S450–S456, 2000.

[23] T. Von Buren, P. D. Mitcheson, T. C. Green, E. M.Yeatman, A. S. Holmes, and G. Troster. Optimizationof inertial micropower generators for human walkingmotion. Sensors Journal, IEEE, 6(1):28–38, 2006.

[24] R. Vullers, R. van Schaijk, I. Doms, C. Van Hoof, andR. Mertens. Micropower energy harvesting. Solid-StateElectronics, 53(7):684–693, 2009.

[25] Y. Wang, M. A. Beydoun, L. Liang, B. Caballero, andS. K. Kumanyika. Will all americans becomeoverweight or obese? estimating the progression andcost of the us obesity epidemic. Obesity,16(10):2323–2330, 2008.

[26] J. Yun, S. N. Patel, M. S. Reynolds, and G. D.Abowd. Design and performance of an optimal inertialpower harvester for human-powered devices. MobileComputing, IEEE Transactions on, 10(5):669–683,2011.